%4d%61%63%68%69%6e%65%4c%65%61%72%6e%69%6e%67
-
KAIST Develops Smartphone AI That Uses Past Solutions to Tackle Similar Problems
A new AI technology has been developed that retains and reuses knowledge, much like writing down a solution in a notebook and applying it to similar problems rather than asking an expert for help each time. Researchers at KAIST have developed a way for small AI models running on smartphones to store and reuse knowledge from a large server model. The approach reduced server calls by an average of 55.61% compared with a non-cumulative approach while maintaining high accuracy, suggesting that mobile AI could make faster decisions with less server support as it encounters similar problems.
KAIST (President Choongsik Bae) announced on September 21 that a research team led by Professor Jae-Gil Lee from the School of Computing has developed CURE (Cumulative Knowledge Reuse), a technology that enables a small AI model on a device to work efficiently with a large model on a server.
Smartphones have limited processing power and memory, so they typically use small, lightweight AI models. These models can handle simple tasks quickly but may be less accurate when identifying complex or unfamiliar images.
Sending every input to a powerful server model can improve accuracy, but transferring data and waiting for a response takes time. It also adds to network traffic and the server’s computational workload, increasing delays for users and operating costs for service providers.
Researchers have explored a collaborative approach in which the on-device model assesses each input first and sends only difficult cases to the server. Without a way to retain the server’s knowledge, however, the device uses each answer once and may need the same help when a similar input appears. This is much like a student who fails to write down the solution to a difficult problem and has to ask for help again when faced with another of the same kind. The team set out to turn these one-time answers into knowledge the device could continue to use.
CURE first checks whether the on-device model can handle an input reliably. If it cannot, the system consults knowledge previously obtained from the server and stored on the device. It contacts the server only if that knowledge is also insufficient. The process follows three steps, moving from solving a problem independently to consulting previous lessons and, when necessary, asking an expert.
For example, an on-device model that cannot identify a car model in a photo may turn to the server for help. CURE uses the server’s prediction and the image’s features to update the device’s knowledge store. When the device later receives a photo of a similar car model, it can use that knowledge to identify it locally.
CURE does not keep a collection of the original photos processed by the server. Instead, it stores a summary of their shared features and differences. This is like noting a car’s distinguishing features, such as its body shape or headlight design, rather than memorizing the entire photo. The stored knowledge can therefore help the system recognize not only images it has already seen but also similar images it encounters for the first time.
The team tested CURE using vision-language models, which connect images with text to understand visual information. These models link what they see to language, much as people do.
The researchers used MobileCLIP2 on the device and EVA-CLIP, which has 18 billion parameters, on the server. Parameters are numerical values within a model that encode information learned during training. They help the model distinguish objects and identify relationships between them.
In tests on a range of image classification datasets, CURE achieved accuracy close to that of an approach that sends every input to the large server model. It also made an average of 55.61 percent fewer server calls than a device-server collaboration baseline that does not retain knowledge from previous server responses.
In end-to-end tests that included communication time, CURE ran up to 2.80 times as fast as the non-cumulative device-server collaboration baseline and up to 3.67 times as fast as the approach that sends every input to the server. Making fewer server calls reduced the time spent transferring data and waiting for responses.
CURE requires no retraining of either the on-device model or the server model. It leaves both models unchanged and adds a separate store for knowledge obtained from the server, making it adaptable to a range of AI models and services.
The technology could reduce repeated data transfers and server computation image recognition. This could mean shorter waits for users and lower operating costs for service providers.
Potential applications also include robots and wearables that need to recognize their surroundings with limited computing resources. Robots operating over slow or unreliable networks, for example, could use previously acquired knowledge to make more decisions locally without waiting for a server response.
The benefits in practice will depend on the device’s processing power and storage capacity, network conditions, and the characteristics of the input data. Further testing across different devices and environments is therefore needed.
“CURE allows a small on-device AI model to remember and reuse knowledge it has already obtained, rather than repeatedly asking the server the same question,” said Professor Jae-Gil Lee. “By maintaining high accuracy while reducing communication demands and response times, we expect it to help smartphones, robots, and wearables use powerful AI models more efficiently.”
Dr. Youngjun Lee, a postdoctoral researcher from the KAIST Institute of Information Electronics, was the study’s first author, and Professor Jae-Gil Lee from the School of Computing was the corresponding author. Co-authors were Doyoung Kim from Amazon, Junhyeok Kang from LG AI Research, and Professor Hwanjun Song from the KAIST Department of Industrial and Systems Engineering. The findings were presented on September 10 at the European Conference on Computer Vision (ECCV 2026), a leading international computer vision conference held in Malmö, Sweden, from September 8 to 12.
Paper title “CURE: Cumulative Knowledge Reuse for Efficient Device-Server Hybrid Inference in Vision-Language Models”,
DOI: https://doi.org/10.1007/978-3-032-37627-5_21
Author information — Dr. Youngjun Lee (postdoctoral researcher, KAIST Institute of Information Electronics, first author), Doyoung Kim (KAIST graduate, now at Amazon, co-author), Junhyeok Kang (KAIST graduate, now at LG AI Research, co-author), Professor Hwanjun Song (KAIST Department of Industrial and Systems Engineering, co-author), Professor Jae-Gil Lee (KAIST School of Computing, corresponding author)
This work was supported by Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. RS-2020-II200862, DB4DL: High-Usability and Performance In-Memory Distributed DBMS for Deep Learning, 50% and No. RS-2022-II220157, Robust, Fair, Extensible Data-Centric Continual Learning, 40%), the InnoCORE pro-gram of the Ministry of Science and ICT (AI Meta-Scientist, N10260110, 10%), and Samsung Electronics Co., Ltd.(IO251216-14634-01).
2026.09.21 View 2752 -
KAIST Develops AI That Can Sense “That’s Not What I Meant” Without Being Told
A future in which AI can recognize a person’s unspoken “that’s not what I meant” response from brain signals and adjust its behavior on its own is coming closer. KAIST researchers have developed a technology that detects cognitive mismatch between humans and AI through brainwaves, enabling AI systems to revise their actions in real time according to human goals. The achievement is expected to accelerate the shift from AI that follows explicit commands to AI that can infer human intent.
KAIST (President Choongsik Bae) announced on September 11 that a research team led by Endowed Chair Professor Sang Wan Lee of the Department of Brain and Cognitive Sciences (Director of the Center for Neuroscience-Inspired Artificial Intelligence), in collaboration with Microsoft Research Asia (MSRA), has developed Neural Value Alignment (NVA), a next-generation brain–computer interface (BCI) technology that uses human brainwaves to align AI behavior with human intentin real time.
For AI to collaborate naturally with people, it must accurately understand what a person actually wants. Until now, however, AI systems have largely inferred human intent from externally observable information such as speech, actions, or gestures.
The challenge is that the same action can reflect different goals, and the same goal can also be achieved through different actions. For example, when a person picks up a cup, it may be unclear whether they intend to drink from it or hand it to someone else. Conversely, if the person's goal is simply to quench their thirst, that same goal could be pursued through several different actions — reaching for the cup, picking up a water bottle, or asking someone else to bring a drink. Because of this two-sided “goal–action ambiguity,” when AI misunderstands human intent, users often have to correct it through additional commands or actions.
To address this problem, the research team focused on the “prediction error” signals that arise unconsciously in the brain when a person encounters an unexpected situation. In simple terms, the team used the brain’s instant “that’s not what I meant” response when AI performs the wrong action or pursues the wrong goal.
The researchers distinguished between two types of brain responses. The first is reward prediction error (RPE), which appears when AI misunderstands the person’s ultimate goal. The second is state prediction error (SPE), which appears when the goal is correct but the process or method of action differs from what the person expected.
The team measured real-time electroencephalography (EEG) signals from people as they observed AI performing tasks. They found that the brain produced different signals depending on whether the AI misunderstood the goal itself or chose the wrong method while pursuing the correct goal. The researchers also identified distinctive brainwave patterns that appeared when both types of errors occurred simultaneously.
By applying deep learning to these brain signals, the team developed a technology that can determine, from EEG alone, how a person is interpreting the AI’s behavior. In other words, even without a person saying “that’s wrong,” the AI can recognize whether the human brain is signaling that “the goal is wrong” or “the method is wrong.”
The team then proposed a Neural Value Alignment-based human–AI synergy algorithm, which feeds these decoded brain signals back to the AI in real time so that it can correct its own behavior.
When the AI detects an SPE signal, it interprets the situation as “the desired goal is correct, but the method is wrong” and adjusts its action strategy. When it detects an RPE signal, it understands that “the goal itself was misunderstood” and searches again for what the person truly intended.
Simulation results showed that the proposed method adapted more quickly than existing approaches even in uncertain situations, such as when a person’s goal suddenly changed or some human neural feedbacks were missing.
The key significance of this research is that it demonstrates the possibility of AI systems correcting themselves by reading a person’s unconscious “that’s not what I meant” brain response, without requiring the user to repeatedly say “do it this way” or “that’s not right.”
As the technology advances, it could be applied to physical AI robots in homes and industrial settings, allowing them to understand user intent more naturally and adjust their actions accordingly. It could also be extended to autonomous vehicles that quickly reflect driver judgment, medical and rehabilitation robots for patients who have difficulty speaking or moving, and educational AI systems that adapt to a student’s cognitive state.
Ultimately, the study presents a new model of human–AI collaboration, moving beyond AI that acts only when explicitly instructed toward AI that can sense human responses and adjust itself accordingly.
Professor Sang Wan Lee, who led the international collaboration, said, “This research is meaningful because it shows that AI can move beyond inferring human intent only from visible behavioral outcomes and instead directly use cognitive signals generated in the brain during collaboration with AI.” He added, “The technology can be expanded to a wide range of fields where human judgment and AI behavior must be closely connected, including physical AI, BCI, autonomous driving, precision personalized education, medical robotics, and human–computer interaction.”
Miran Lee, Director, Microsoft Research Accelerator at Microsoft Research, said, “This achievement is the result of the ongoing international collaboration between KAIST and Microsoft Research Asia. We look forward to continuing this partnership to develop world-class BCI technologies that enable humans and AI to communicate and collaborate more naturally.”
The study’s first author is Xin Xu, a Ph.D. student in KAIST’s Department of Brain and Cognitive Sciences. Researchers from Microsoft Research Asia, including Yansen Wang, Dongqi Han, and Dongsheng Li, also participated in the study. The research was published online in August 2026 in IEEE Transactions on Cybernetics, an international journal in the field of cybernetics.
Paper title: Neural Value Alignment: Human–AI Collaboration Under Goal–Action Ambiguity
DOI: 10.1109/TCYB.2026.3722605
Another KAIST–Microsoft Research Asia collaborative study on helping AI rapidly adapt to continuously changing environments was presented in June at ICML 2026, one of the leading international conferences in artificial intelligence. The study was led by Niklas Koeppe, a Ph.D. student in KAIST’s Program of Brain and Cognitive Engineering, as first author.
Paper title: Mitigating Plasticity Loss through Architectural Design in Continual Learning
Original paper: https://icml.cc/virtual/2026/poster/61534
This research was supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP), funded by the Ministry of Science and ICT.
2026.09.10 View 3516